420 / 1020
ESTRO 35 2016 S397
________________________________________________________________________________
of automatic (AU) and manual (MA) generated H&N VMAT
plans created for clinical use.
Material and Methods:
All patients (n=30) referred to
curative H&N radiotherapy in August and September 2015
were planned with a MA and AU VMAT plan in Pinnacle
version 9.10. Half of the tumours were located in the pharynx
(15) and the rest were mixed between larynx (4), oral cavity
(3), salivary glands (2), thyroid gland and unknown primary
(4). The plans followed national guidelines, and planning
techniques were blinded before clinical evaluation of senior
oncologists. The MA plans were optimized according to
standard clinical practice. The AU plans were created by the
Autoplan software available in the Pinnacle planning system.
After AU optimisation, slight manual fine-tuning of the plans
was performed. To supplement the clinical evaluation the
operator time for the dosimetrist was recorded along with
DVH parameters and the global detector pass rate (3% and
3mm) of the delivered plans on an ArcCheck phantom. All
statistical analyses were performed with a paired Wilcoxon-
signed rank test.
Results:
In 29/30 plans, the AU plan was chosen for clinical
application (p<0.001). In terms of DVH parameters similar
target coverage was obtained between the two planning
methods. As seen in the table, mean OAR doses were
significantly reduced in the AU plans for all organs. The mean
reduction ranged from 0.5 Gy for the entire patient to 6.5 Gy
for the contralateral submandibular gland. Differences in DVH
showed significant AU superiority in the dose range 10 Gy to
45 Gy for all organs (Mean DVH example shown in figure). The
only manual plan selected for clinical use was a thyroid
cancer plan involving level VII lymph nodes and therefore
included a large volume of the lung, which had a lower lung
dose in the manual plan. The AU plans were more modulated
as illustrated by the increase in MUs, which might cause the
reduced, but still clinically acceptable, pass rate of 97.7% in
ArcCheck measurements. The increased beam-on time of 4
sec is clinically unimportant. Mean operator time spent on MA
plans was more than twice that of AU plans. The target
homogeneity, conformity and dose fall off were all superior
in the AU plans.
Conclusion:
All AU and MA plans were of acceptable clinical
quality, however, AU plans resulted in reduced doses to all
OAR and required less operator time in the planning process.
AU plans were almost consistently preferred by senior head
and neck cancer specialists. The dosimetric superiority of the
AU plans was evident.
PO-0838
Impact of dosimetric outliers on the performance of a
knowledge-based planning system
A. Delaney
1
VUMC, Radiotherapy, Amsterdam, The Netherlands
1
, J. Tol
1
, M. Dahele
1
, J. Cuijpers
1
, B. Slotman
1
, W.
Verbakel
1
Purpose or Objective:
RapidPlan (Varian Medical Systems) is
a knowledge-based planning solution based on a model
derived from a library of previous treatment plans. The
model utilizes the geometric features and associated
dosimetry of these plans to predict a range of achievable
dose volume histograms (DVHs) for each organ at risk (OAR)
of a new patient. RapidPlan (RP) drives the VMAT or IMRT
optimization process by placing a line of optimization
objectives along the lower boundary of the DVH prediction
range. Planning inconsistencies may lead to sub-optimal plans
in the model, which can be identified as dosimetric outliers.
Outlier cleaning is advised, however this is time consuming
and often subjective. We examined the effect of model
cleaning and increasing numbers of dosimetric outliers in a
model library, on RP plan quality.
Material and Methods:
70 head and neck cancer treatment
plans (planned consistently using the same departmental
objective priorities) were used to populate uncleaned
ModelUC. Statistical metrics provided by RP were used to
identify geometric/dosimetric outliers in ModelUC which
were then visually assessed and, if appropriate, removed to
create cleaned ModelC. The last 5-40 patients (increments of
5) of ModelC were then re-planned with no attempt to spare
the salivary glands, and afterward re-introduced to ModelC,
creating Model5-40. All models were used to create plans for
a 10 patient evaluation group. Although RP can generate OAR
objective priorities, for this study, the same standard
priorities were used as for the plans in ModelUC. Plans were
compared on the basis of generated prediction ranges,
boost/elective target volume homogeneity index (HIB/HIE),
mean dose to the oral cavity (OC) and to composite
structures comprising the salivary glands (compsal) and
swallowing structures (compswal).